Re: [savi] About draft-dong-savi-cga-header-02.txt

[Alberto García <[e-mail]>]

Hi Dong,

|  -----Mensaje original-----
|  De: [e-mail] [mailto:[e-mail]] En nombre de Dong
|  Zhang
|  Enviado el: jueves, 23 de julio de 2009 5:04
|  Para: Alberto_García; 'Padmanabha Nallur'
|  CC: 'SAVI Mailing List'; 'Christian Vogt'
|  Asunto: Re: [savi] About draft-dong-savi-cga-header-02.txt
|  
|  Hi Alberto,
|  Alberto_García 2009-07-22 Wrote:
|  >Hi,
|  >
|  >|  -----Mensaje original-----
|  >|  De: Padmanabha Nallur [mailto:[e-mail]]
|  >|  Enviado el: lunes, 20 de julio de 2009 23:01
|  >|  Para: 'Alberto García'
|  >|  CC: 'Dong Zhang'; 'Christian Vogt'; 'SAVI Mailing List'
|  >|  Asunto: RE: [savi] About draft-dong-savi-cga-header-02.txt
|  >|
|  >|  Hi Alberto,
|  >|
|  >|  The main point of the CGA proposal is to avoid IP spoofing.
|  >|  In order to avoid IP spoofing the source IP address checks must be
|  performed
|  >|  closer to the destination or destination network as opposed to implementing
|  >|  these checks in the source network (e.g., SAVI).
|  >|
|  >|  The CGA proposal provides an alternative solution to prevent IP spoofing by
|  >|  providing the checks that can be implemented in the destination endpoint or
|  >|  destination network (actually anywhere in the network). We would like to
|  >|  have as simple solution as possible and preferably even stateless solution.
|  >
|  >In my opinion, what you are proposing is dangerous. Let's consider flood-based
|  DoS (draft-ietf-savi-threat-scope-00.txt): an attacker wants to exhaust the
|  victim's resources by sending traffic to it. With your mechanism, packets with
|  false CGA information do not only consume TCP state, memory for IP
|  fragmentation... but they also force the host to spend a lot of CPU in using public
|  cryptography to check a CGA! This is what any attacker would like: force hosts to
|  do a lot of work for fake packets!
|  >This is why in Shim6 and HIP a 4-way handshake is performed: the initiator
|  must prove at least that the address is using is topologically valid (for accounting
|  purposes).
|  We are aware of the cost of the computation and the a few possible DoS attacks.
|  Please read the section 8 and the security consideration in the draft. Would these
|  remove your worry?

Not really. 
As far as I understand, the three types of verification you propose (8.1, 8.2, 8.3) are allowed. 
If you allow 8.1, an host trying to DoS a victim's node will send a CGA Request, and the victim's node will answer with 'CGA Params, CGA Sig'. This costs the victim some CPU: the CGA Request includes a sequence number, and the CGA signature includes this sequence number, so it has to compute specifically for every packet sent.
To protect from this you say in 10: "When one host receives packets with CGA Request from a same source address repeatedly, it MUST refuse to return the CGA Params and CGA Signature." But, of course, a single attacker is going to use different addresses every time!  
So, IMHO, 8.1 introduces a new vulnerability that was not present before. You need to rate-limit the generation of this in general, not for any specific source, and then you should accept packets not protected.
So, if a node wants to impersonate another, first sends lots of requests to start the rate-limited mode, and then sends unprotected packets from a source to impersonate. And the purpose of the mechanism is defeated.

8.3 is almost Shim6. But, since with Shim6 a state is created, you don't need to convey the CGA information on every packet. Security is less (you are not requesting all the time the other end to prove it has the private key), but it may be enough for this purpose, since it was proved at the beginning the address ownership, and the will of the remote host to communicate. 
This is statefull, and intermediate routers could not perform any kind of filtering based on this information unless they trace the whole exchange. 
But I don't see which protection could be provided without state: you have included a sequence number and a request/response mechanism that use it (8.1, 8.2, 8.3), and it is a good approach because it protects against reply attacks. If intermediate routers are not aware of this exchange (because they work without state), they are prone to reply attacks. 
I really don't see a clear benefit/cost/security_risk balance for allowing routers in any intermediate location to perform such filtering.

8.2 is an in-between 8.1 - 8.3

Regards,
Alberto

|  
|  Thanks.
|  >
|  >In addition, you propose not only perform this DoS-prone check not only in the
|  host, but  anywhere in the network. So any router should divert a packet from
|  the silicon fast path to the router CPU to check the CGA... That would be double
|  prize for the attacker: not only compromising a host, but the whole
|  communication pipe of a site.
|  >
|  >I really don't see this.
|  >
|  >Regards,
|  >Alberto
|  >
|  >|  This is the main goal of CGA proposal and it would be great if it can ALSO
|  >|  be used in other situations (e.g., SAVI) as you mentioned below.
|  >|
|  >|  Validation of the IP address itself (like prefix check) is not the main goal
|  >|  of the CGA proposal.
|  >|
|  >|
|  >|  Thanks
|  >|  Paddy Nallur
|  >|
|  >|  -----Original Message-----
|  >|  From: Alberto García [mailto:[e-mail]]
|  >|  Sent: Monday, July 20, 2009 3:29 AM
|  >|  To: 'Dong Zhang'; 'Christian Vogt'; 'SAVI Mailing List'
|  >|  Cc: 'Padmanabha Nallur 73868'
|  >|  Subject: RE: [savi] About draft-dong-savi-cga-header-02.txt
|  >|
|  >|  Hi
|  >|
|  >|  |  -----Mensaje original-----
|  >|  |  De: [e-mail] [mailto:[e-mail]] En nombre de
|  >|  Dong
|  >|  |  Zhang
|  >|  |  Enviado el: lunes, 20 de julio de 2009 4:41
|  >|  |  Para: Alberto_García; 'Christian Vogt'; 'SAVI Mailing List'
|  >|  |  CC: 'Padmanabha Nallur 73868'
|  >|  |  Asunto: Re: [savi] About draft-dong-savi-cga-header-02.txt
|  >|  |
|  >|  |  Hi Alberto,
|  >|  |  Please check inline.
|  >|  |
|  >|  |  Thank you.
|  >|  |
|  >|  |  Alberto_García 2009-07-17 Wrote:
|  >|  |  >Hi
|  >|  |  >
|  >|  |  >|  -----Mensaje original-----
|  >|  |  >|  De: [e-mail] [mailto:[e-mail]] En nombre
|  de
|  >|  |  Dong
|  >|  |  >|  Zhang
|  >|  |  >|  Enviado el: viernes, 17 de julio de 2009 8:50
|  >|  |  >|  Para: Alberto_Garc?; 'Christian Vogt'; 'SAVI Mailing List'
|  >|  |  >|  CC: 'Padmanabha Nallur 73868'
|  >|  |  >|  Asunto: Re: [savi] About draft-dong-savi-cga-header-02.txt
|  >|  |  >|
|  >|  |  >|  Alberto_Garc韆 2009-07-16 Wrote:
|  >|  |  >|  >|  I do not think it is a conflict. The difference is just how to
|  >|  |  >|  >|  use CGA. For instance, the CGA proposal can be used in e2e, even
|  >|  |  between
|  >|  |  >|  >the
|  >|  |  >|  >|  host
|  >|  |  >|  >|  and the first-hop router. It can depend on the destination option
|  >|  header.
|  >|  |  >|  >
|  >|  |  >|  >Shim6 allows transporting CGA information end-to-end. It does that
|  >|  now, in
|  >|  |  a
|  >|  |  >|  >standard way. Why you should want to generate a yet-another-
|  protocol
|  >|  to
|  >|  |  >|  >transport CGA information end-to-end?
|  >|  |  >|  Thank you for the information.
|  >|  |  >|  We will read Shim6. If it really does the relative job, we will be
|  >|  glad to use it.
|  >|  |  >|  >
|  >|  |  >|  >Besides, what does such a protocol would have to do with SAVI?
|  >|  |  >|  >If you want to use CGAs for providing security in the link of the
|  >|  host and
|  >|  |  >|  >the first-hop router, you should probably use the SEND protocol (see
|  >|  |  >|  >draft-ietf-savi-send-00). Do you think there is something in the
|  >|  SEND
|  >|  |  >|  >protocol that does not suit with binding IPv6 addresses and link
|  >|  layer
|  >|  |  >|  >anchors, that could be improved?
|  >|  |  >|  >And if you want to use the CGA for address validation further than
|  >|  the first
|  >|  |  >|  >hop routers, I see some reasons why you should not:  at that point
|  >|  you
|  >|  |  need
|  >|  |  >|  >to validate prefixes, not host addresses, and CGA provides no
|  >|  protection for
|  >|  |  >|  >prefixes: any host can generate a valid CGA with whichever prefix it
|  >|  wanted.
|  >|  |  >|  >So just validating a CGA does not assure that the host is authorized
|  >|  to use
|  >|  |  >|  >that prefix.
|  >|  |  >|  In the CGA-related information, it contains the signature. We think
|  >|  it can
|  >|  |  >|  protect the prefix. Right?
|  >|  |  >
|  >|  |  >Well, it does not prove that a host has been legitimately assigned a
|  >|  prefix.
|  >|  |  >CGAs, with a protocol transporting signatures made with the private key,
|  >|  only
|  >|  |  assure that once a host generates a given address, it is very very
|  >|  difficult for
|  >|  |  another host to use the same address. And yes, the protection includes
|  >|  the
|  >|  |  prefix: it is just not enough to find a key pair whose hash generates a
|  >|  given
|  >|  |  interface identifier, and then you can impersonate any host with the same
|  >|  |  interface identifier in any network (i.e. with any prefix).
|  >|  |  [Dong] Well, yes, we agree with this. CGA does not do identity
|  >|  authentication.
|  >|  |  We only validate the address so that someone else cannot use this IP
|  >|  address.
|  >|  |  We do not check for the IP address binding with a host  or a user or some
|  >|  other
|  >|  |  higher level entity and validate that.
|  >|  |
|  >|  |  Just the fact that someone else cannot use the same address would give
|  >|  the
|  >|  |  benefits mentioned in the use cases within the draft.
|  >|
|  >|  I think the place to do this is in the first link. And the tool SEND + SAVI.
|  >|  Outside the first link, I think it is too costly to do checks in a
|  >|  per-address basis, and you need other tools (manual configuration, routing
|  >|  info...) for validating prefixes... so just validate prefixes with this
|  >|  tools.
|  >|
|  >|  |
|  >|  |  IMHO, we don't think even SAVI solutions need to validate the binding of
|  >|  IP
|  >|  |  address with users etc., only check if a certain IP address is used on a
|  >|  specific
|  >|  |  link.
|  >|
|  >|  Yes, of course. I was not talking at any point in "users".
|  >|  Again, to check the address, use SEND.
|  >|
|  >|  |  >
|  >|  |  >But that says nothing about the legitimacy of ah host to use that
|  >|  prefix.
|  >|  |  >For example, give the prefix of your network, and I can generate
|  >|  immediately
|  >|  |  an CGA that claims to belong to your prefix: CGA = hash ( prefix | public
|  >|  key |
|  >|  |  etc). And I can use from my network, if the router (and some routers
|  >|  above)
|  >|  |  does not check source prefixes. Looking only at the CGA nobody can know
|  >|  if my
|  >|  |  host is authorized to use a prefix or not.
|  >|  |  >So, CGA does not add anything to prefix protection. You have to provide
|  >|  other
|  >|  |  means to protect that.
|  >|  |  I guess the ingress filtering could do this. And savi solutions are the
|  >|  complement
|  >|  |  for ingress filtering. This point has been discussed before.
|  >|  |  >
|  >|  |  >
|  >|  |  >|  >
|  >|  |  >|  >|  >
|  >|  |  >|  >|  >What does this type of solutions has to do with SAVI,
|  >|  considering that
|  >|  |  >|  >the
|  >|  |  >|  >|  >charter says: "No changes to hosts are allowed."? Isn't your
|  >|  solution
|  >|  |  >|  >|  >proposing a change to hosts?
|  >|  |  >|  >|  Yes, the CGA proposal does not fit for the current charter quit
|  >|  well to
|  >|  |  >|  >some
|  >|  |  >|  >|  extend.
|  >|  |  >|  >|  The reason why we put the CGA-related info in the extension
|  >|  header is
|  >|  |  >|  >that it
|  >|  |  >|  >|  has
|  >|  |  >|  >|  the ability to provide a basic security for the source address
|  >|  like AH &
|  >|  |  >|  >ESP. The
|  >|  |  >|  >
|  >|  |  >|  >The CGA security model does not have much relationship with AH &
|  >|  ESP.
|  >|  |  Well,
|  >|  |  >|  >all use keys, but besides that...
|  >|  |  >|  I mean if we could take the CGA-related information in the extension
|  >|  header,
|  >|  |  >|  the packet will have AH, ESP and CGA for the basic security
|  >|  characteristics,
|  >|  |  >|  including integrity, confidentiality, and authentication.
|  >|  |  >
|  >|  |  >This is completely different stuff. It seems like generating an
|  >|  opportunistic IKE
|  >|  |  security context (a la SSL) for a communication, starting with the
|  >|  public/private
|  >|  |  key pair associated to the CGAs of two hosts. Then you should look at
|  >|  |  http://tools.ietf.org/id/draft-laganier-ike-ipv6-cga-02.txt
|  >|  |  I am a little confused here. I have already known this draft before. Are
|  >|  you
|  >|  |  comparing it with my another draft,
|  >|  http://tools.ietf.org/id/draft-zhang-shen-sa-
|  >|  |  cga-00.txt ?
|  >|
|  >|  I was not aware of your draft. I wasn't comparing anything, but, from the
|  >|  title of your draft, yes it seems similar.
|  >|
|  >|  Regards,
|  >|  Alberto
|  >|
|  >|  |
|  >|  |
|  >|  |  ------------------
|  >|  |  Dong Zhang
|  >|  |  2009-07-20
|  >
|  >.
|